content-recommendation
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input from ServiceNow records to perform searches and update work notes, creating a surface for indirect instructions.\n
- Ingestion points: Processes content from
incident(short_description, description),sn_customerservice_case, andkb_feedback(comments) tables.\n - Boundary markers: The skill does not implement delimiters or explicit instructions to distinguish between data and agent commands when building search queries or record updates.\n
- Capability inventory: The skill possesses the capability to modify ServiceNow records via
SN-Update-RecordandSN-Add-Work-Notes.\n - Sanitization: No evidence of sanitization or structural validation is present for the data ingested from ServiceNow before it is used in prompt context or written back to the system.
Audit Metadata